Multiple Databases

DETAILS: Tier: Free, Premium, Ultimate Offering: Self-managed

Introduced in GitLab 15.7.

WARNING: This feature is not ready for production use

By default, GitLab uses a single application database, referred to as the main database.

To scale GitLab, you can configure GitLab to use multiple application databases.

Due to known issues, configuring GitLab with multiple databases is in limited beta.

After you have set up multiple databases, GitLab uses a second application database for CI/CD features, referred to as the ci database. We do not exclude hosting both databases on a single PostgreSQL instance.

All tables have exactly the same structure in both the main, and ci databases. Some examples:

When multiple databases are configured, the ci_pipelines table exists in both the main and ci databases, but GitLab reads and writes only to the ci_pipelines table in the ci database.
Similarly, the projects table exists in both the main and ci databases, but GitLab reads and writes only to the projects table in the main database.
For some tables (such as loose_foreign_keys_deleted_records) GitLab reads and writes to both the main and ci databases. See the development documentation

Known issues

Once data is migrated to the ci database, you cannot migrate it back.
Significant downtime is expected for larger installations (database sizes of more 100 GB).
Running two databases is not yet supported with Geo.

Migrate existing installations using a script

A script for migrating existing Linux package installations was introduced in GitLab 16.8.

Existing Linux package installations

This migration requires downtime. If something unexpected happens during the migration, it is safe to start over.

Preparation

Verify available disk space:
- The database node that will store the gitlabhq_production_ci database needs enough space to store a copy of the existing database: we duplicate gitlabhq_production. Run the following SQL query to find out how much space is needed. Add 25%, to ensure you will not run out of disk space.
```
sudo gitlab-psql -c "SELECT pg_size_pretty( pg_database_size('gitlabhq_production') );"
```
- During the process, a dump of the gitlabhq_production database needs to be temporarily stored on the filesystem of the node that will run the migration. Execute the following SQL statement to find out how much local disk space will be used. Add 25%, to ensure you will not run out of disk space.
```
sudo gitlab-psql -c "select sum(pg_table_size(concat(table_schema,'.',table_name))) from information_schema.tables where table_catalog = 'gitlabhq_production' and table_type = 'BASE TABLE'"
```
Plan for downtime. The downtime is dependent on the size of the gitlabhq_production database.
- We dump gitlabhq_production and restore it into a new gitlabhq_production_ci database. Database sizes below 50 GB should be done within 30 minutes. Larger databases need more time. For example, a 100 GB database needs 1-2 hours to be copied to the new database.
- We advise to also plan some time for smaller tasks like modifying the configuration.

Create the new gitlabhq_production_ci database:

sudo gitlab-psql -c "CREATE DATABASE gitlabhq_production_ci WITH OWNER 'gitlab'"

Migration

This process includes downtime. Running the migration script will stop the GitLab instance. After the migration has been finished, the instance is restarted.

Create a backup of the configuration:

sudo cp /etc/gitlab/gitlab.rb /etc/gitlab/gitlab.rb.org

Edit /etc/gitlab/gitlab.rb and save the changes. Do not run the reconfigure command, the migration script will run that for you.

gitlab_rails['env'] = { 'GITLAB_ALLOW_SEPARATE_CI_DATABASE' => 'true' }
gitlab_rails['databases']['ci']['enable'] = true
gitlab_rails['databases']['ci']['db_database'] = 'gitlabhq_production_ci'

Run the migration script:

sudo gitlab-ctl pg-decomposition-migration

At this point, the GitLab instance should start and be functional.

If you want to abort the procedure and you want to start GitLab without changing anything, run the following commands:

sudo cp /etc/gitlab/gitlab.rb.org /etc/gitlab/gitlab.rb
sudo gitlab-ctl reconfigure
sudo gitlab-ctl restart

Cleaning up

If everything works as expected, we can clean up unneeded data:

Delete the CI data in Main database:

sudo gitlab-rake gitlab:db:truncate_legacy_tables:main

Delete the Main data in CI database:

sudo gitlab-rake gitlab:db:truncate_legacy_tables:ci

Migrate existing installations (manual procedure)

To migrate existing data from the main database to the ci database, you can copy the database across.

NOTE: If something unexpected happens during the migration, it is safe to start over.

Existing self-compiled installation

Disable background migrations.
Ensure all background migrations are finished.

Stop GitLab, except for PostgreSQL:

sudo service gitlab stop
sudo service postgresql start

Dump the main database:

sudo -u git pg_dump -f gitlabhq_production.sql gitlabhq_production

Create the ci database, and copy the data from the previous dump:

sudo -u postgres psql -d template1 -c "CREATE DATABASE gitlabhq_production_ci OWNER git;"
sudo -u git psql -f gitlabhq_production.sql gitlabhq_production_ci

Configure GitLab to use multiple databases.

Existing Linux package installations

Disable background migrations
Ensure all background migrations are finished

Stop GitLab, except for PostgreSQL:

sudo gitlab-ctl stop
sudo gitlab-ctl start postgresql

Dump the main database:

sudo -u gitlab-psql /opt/gitlab/embedded/bin/pg_dump -h /var/opt/gitlab/postgresql -f gitlabhq_production.sql gitlabhq_production

Create the ci database, and copy the data from the previous dump:

sudo -u gitlab-psql /opt/gitlab/embedded/bin/psql -h /var/opt/gitlab/postgresql -d template1 -c "CREATE DATABASE gitlabhq_production_ci OWNER gitlab;"
sudo -u gitlab-psql  /opt/gitlab/embedded/bin/psql -h /var/opt/gitlab/postgresql -f gitlabhq_production.sql gitlabhq_production_ci

Configure GitLab to use multiple databases.

Existing Linux package installations using streaming replication

To reduce downtime, you can set up streaming replication to migrate existing data from the main database to the ci database. This procedure results in two database clusters.

This procedure can be both time- and resource-consuming. Consider their trade-offs with availability before executing it.

To set up streaming replication for creating two database clusters:

Set up streaming replication from the GitLab database to new database instance.
When the new replica has caught up, disable background migrations.
Ensure all background migrations are finished.

Stop GitLab, except for PostgreSQL:

sudo gitlab-ctl stop
sudo gitlab-ctl start postgresql

After the replication is complete, stop the streaming replication, and promote the replica to a primary instance. You now have two database clusters, one for main, and one for ci.
Configure GitLab to use multiple databases.

For more information on how to set up Streaming Replication, see PostgreSQL replication and failover for Linux package installations.

Set up multiple databases

To configure GitLab to use multiple application databases, follow the instructions below for your installation type.

WARNING: You must stop GitLab before setting up multiple databases. This prevents split-brain situations, where main data is written to the ci database, and the other way around.

Self-compiled installations

For existing installations, migrate the data first.
Back up GitLab in case of unforeseen issues.
Stop GitLab:
```
sudo service gitlab stop
```
Open config/database.yml, and add a ci: section under production:. See config/database.yml.decomposed-postgresql for possible values for this new ci: section. Once modified, the config/database.yml should look like:
```
production:
  main:
    # ...
  ci:
    adapter: postgresql
    encoding: unicode
    database: gitlabhq_production_ci
    # ...
```
Save the config/database.yml file.
Update the service files to set the GITLAB_ALLOW_SEPARATE_CI_DATABASE environment variable to true.

For new installations only. Create the gitlabhq_production_ci database:

sudo -u postgres psql -d template1 -c "CREATE DATABASE gitlabhq_production OWNER git;"
sudo -u git -H bundle exec rake db:schema:load:ci

Lock writes for ci tables in main database, and the other way around:
```
sudo -u git -H bundle exec rake gitlab:db:lock_writes
```
Restart GitLab:
```
sudo service gitlab restart
```
Enable background migrations

Linux package installations

For existing installations, migrate the data first.
Back up GitLab in case of unforeseen issues.
Stop GitLab:
```
sudo gitlab-ctl stop
```

Edit /etc/gitlab/gitlab.rb and add the following lines:

gitlab_rails['env'] = { 'GITLAB_ALLOW_SEPARATE_CI_DATABASE' => 'true' }
gitlab_rails['databases']['ci']['enable'] = true
gitlab_rails['databases']['ci']['db_database'] = 'gitlabhq_production_ci'

Save the /etc/gitlab/gitlab.rb file.
Reconfigure GitLab:
```
sudo gitlab-ctl reconfigure
```

Optional, for new installations only. Reconfiguring GitLab should create the gitlabhq_production_ci database if it does not exist. If the database is not created automatically, create it manually:

sudo gitlab-ctl start postgresql
sudo -u gitlab-psql /opt/gitlab/embedded/bin/psql -h /var/opt/gitlab/postgresql -d template1 -c "CREATE DATABASE gitlabhq_production_ci OWNER gitlab;"
sudo gitlab-rake db:schema:load:ci

Lock writes for ci tables in main database, and the other way around:

sudo gitlab-ctl start postgresql
sudo gitlab-rake gitlab:db:lock_writes

Restart GitLab:
```
sudo gitlab-ctl restart
```
Enable background migrations

Further information

For more information on multiple databases, see issue 6168.

For more information on how multiple databases work in GitLab, see the development guide for multiple databases.

Since 2022-07-02, GitLab.com has been running with two separate databases. For more information, see this blog post.